<b>Statistics-Enhanced Cross-Domain Perception Network for Spaceborne Infrared Tiny Ship Detection</b>

Our SDG-TISD dataset is constructed from real infrared remote sensing imagery acquired by the Thermal Infrared Spectrometer (TIS) onboard the SDGSAT-1 satellite. SDGSAT-1 is the world’s first scientific satellite designed for the United Nations Sustainable Development Goals (SDGs). It is equipped with a Multispectral Imager (MII), a high-resolution wide-swath Thermal Infrared Spectrometer (TIS), and a low-light-level Grating Imager Unit (GIU), enabling all-weather, day-and-night, and multi-payload synergistic observations. Among these, TIS is the world’s first civilian thermal infrared spectrometer providing three spectral bands, with a spatial resolution of 30 m and a swath width of 300 km.To construct SDG-TISD, we first trained a YOLOv9s detector on a small set of carefully annotated samples provided by human experts. This preliminary model was then applied to all TIS imagery collected between July 2022 and April 2024 to generate candidate ship detections. Each candidate was subsequently cross-validated with vessel records from the Automatic Identification System (AIS) database provided by MarineTraffic, using both temporal and geospatial consistency as matching criteria. Detections with perfect spatio-temporal correspondence to AIS data were considered true positives, while unmatched ships (e.g., those not updating AIS information, illegal vessels, or errors introduced by the pre-detection model) were carefully inspected and re-annotated.After rigorous screening and refinement, the final SDG-TISD dataset contains 7,660 image chips of size 256 × 256 pixels, with a total of 12,073 ship instances. The dataset provides thermal infrared imagery at a 30 m spatial resolution, covering the three TIS spectral bands (8–10.5 μm, 10.3–11.3 μm, and 11.5–12.5 μm).